Driver Assistant iOS App

The Challenge

Building a robust driver assistant application requires highly accurate, real-time object detection capable of running efficiently on a mobile device. This involved overcoming several hurdles: training a specialized object detection model, optimizing it for on-device performance (given mobile resource constraints), and integrating it seamlessly into an iOS application with real-time data from the device's sensors. The goal was to reliably identify critical traffic elements and provide immediate feedback to the user.

My Solution

This project involved an end-to-end development pipeline. I started by creating COCO Traffic, a custom dataset specifically designed to recognize objects commonly encountered in traffic, such as pedestrians, red traffic lights, and stop signs. I then trained a YOLOv5s object detector on this dataset and subsequently optimized the model for deployment on iOS devices using Core ML.

The core of the solution was developing the iOS application in Swift. This involved:

• Integrating the optimized ML model to perform on-device object detection from the camera feed.
• Implementing logic to highlight critical objects (pedestrians, red traffic lights, stop signs) visually.
• Calculating and displaying the current speed using GPS data for real-time driver awareness.
• Fine-tuning the model and app for optimal performance on mobile devices, ensuring smooth operation and responsiveness.

The project demonstrates a complete cycle from custom dataset creation and model training to mobile app development and on-device machine learning optimization.

The Outcome

• Real-time On-Device AI: Successfully deployed a powerful object detection model capable of high-performance, real-time inference directly on an iOS device, eliminating cloud dependency.
• Custom Data & Model Expertise: Demonstrated the ability to create specialized datasets and train tailored deep learning models for specific real-world applications.
• End-to-End Mobile AI Development: Showcased comprehensive skills in data preparation, ML model training, optimization for mobile, and native iOS application development (Swift).
• Enhanced Driver Awareness: Provided a functional tool that aids drivers by visually highlighting potential hazards and displaying crucial speed information in real-time.